Wine chemical markers assess nitrogen levels in original grape juice

Abstract

Nitrogen (N) nutrition of the vineyard plays a crucial role in the composition of must and wine, impacting fermentation, as well as the aroma and taste of the final product. N-deficient grape juice can result in increased astringency and bitterness, and a decrease in pleasant aromas in the wine. N management in vineyards is continually evolving, influenced by climate change and emerging trends in cover crop management. These factors can affect the availability of N to the vines. Yeast-assimilable nitrogen (YAN) in grape juice is a reliable indicator of the N status of vines. Ideally, YAN should be measured at harvest to identify deficiency (YAN < 140 mg/L). However, this practice is not widely adopted, and once the wine is produced, the original YAN levels in the must cannot be determined.

This study proposes a methodology to estimate YAN concentrations in the original grape juice by analysing the wine. Several chemical markers found in wine have been identified as potential indicators of N deficiency in the grape must for the Chasselas cultivar [1]. We suggest using a predictive model based on four of these markers: proline, succinic acid, 2-phenylethanol (PhEtOH), and 2,3-methylbutanol. These markers are known to be present in all grape varieties and remain stable during wine aging.

The study builds several predictive models: a linear model as a baseline, a generalized additive model to handle non-linear relationships, and a random forest model (a flexible machine learning algorithm). We assess their predictive power using a test set (data not used in the training process). The dataset includes results from grape juice and wine analyses of 447 wines from 16 grape varieties grown in Switzerland by Agroscope between 2014 and 2023. The model provides an acceptable estimation of YAN deficiency across all grape varieties. When a single grape variety with a reasonable sample size (129) is considered, the estimation is improved to reach a median relative absolute error of 8.6% (meaning that 50 % of predictions fall within an interval of the observed value ± 8.6 %). The predictive analyses suggest that the markers with highest predictive power are the proline and PhEtOH.

This methodology has the potential to help winegrowers monitor N status post-fermentation and adjust vineyard practices accordingly, leading to improvements in wine quality. In the future, a possible web app’ allowing winemakers to make one’s own prediction is envisioned.

References

[1] Dienes-Nagy, Á., Marti, G., Breant, L., Lorenzini, F., Fuchsmann, P., Baumgartner, D., Zufferey, V., Spring, J-L., Gindro, K., Viret, O., Wolfender, J-L., Johannes Rösti, J. OENO One (2020), 54(3), 583–599.